The branch main has been updated by fuz:
URL:
https://cgit.FreeBSD.org/src/commit/?id=127709d30a2b8a38be995dc5053390947895a723
commit 127709d30a2b8a38be995dc5053390947895a723
Author: Robert Clausecker <f...@freebsd.org>
AuthorDate: 2024-11-18 12:30:55 +0000
Commit: Robert Clausecker <f...@freebsd.org>
CommitDate: 2024-12-02 10:41:11 +0000
lib/libc/gen: use Lemire's algorithm for arc4random_uniform().
Daniel Lemire has published a more efficient range reduction algorithm
for finding a random number in a given range without bias, reducing the
number of divisions to none in the common case and 1 in case the initial
sample is rejected.
This speeds up performance by 22% on amd64, 15% on i386, and 70% on armv7.
os: FreeBSD
arch: amd64
cpu: Intel(R) Core(TM) i7-4910MQ CPU @ 2.90GHz
│ benchmark.out │
│ sec/op │
Arc4random_uniform 56.53n ± 0%
Fast_uniform 44.00n ± 0%
geomean 49.87n
Reviewed by: cem
Approved by: emaste
Differential Revision: https://reviews.freebsd.org/D47659
---
lib/libc/gen/arc4random.3 | 13 ++++++-
lib/libc/gen/arc4random_uniform.c | 71 +++++++++++++++++----------------------
2 files changed, 43 insertions(+), 41 deletions(-)
diff --git a/lib/libc/gen/arc4random.3 b/lib/libc/gen/arc4random.3
index 1d5b3b9f43cb..1b042f15f000 100644
--- a/lib/libc/gen/arc4random.3
+++ b/lib/libc/gen/arc4random.3
@@ -30,7 +30,7 @@
.\"
.\" Manual page, using -mandoc macros
.\"
-.Dd April 13, 2020
+.Dd November 18, 2024
.Dt ARC4RANDOM 3
.Os
.Sh NAME
@@ -129,6 +129,17 @@ functions using
.%O Document ID: 4027b5256e17b9796842e6d0f68b0b5e
.%U http://cr.yp.to/papers.html#chacha
.Re
+.Rs
+.%A Daniel Lemire
+.%T Fast Random Integer Generation in an Interval
+.%D January 2019
+.%J ACM Trans. Model. Comput. Simul.
+.%I Association for Computing Machinery
+.%C New York, NY, USA
+.%V vol. 29
+.%N no. 1
+.%P pp. 1\(en12
+.Re
.Sh HISTORY
These functions first appeared in
.Ox 2.1 .
diff --git a/lib/libc/gen/arc4random_uniform.c
b/lib/libc/gen/arc4random_uniform.c
index 06cd29c6dbe4..23455e545899 100644
--- a/lib/libc/gen/arc4random_uniform.c
+++ b/lib/libc/gen/arc4random_uniform.c
@@ -1,56 +1,47 @@
-/* $OpenBSD: arc4random_uniform.c,v 1.3 2019/01/20 02:59:07 bcook Exp $
*/
-
-/*
- * Copyright (c) 2008, Damien Miller <d...@openbsd.org>
+/*-
+ * SPDX-License-Identifier: 0BSD
*
- * Permission to use, copy, modify, and distribute this software for any
- * purpose with or without fee is hereby granted, provided that the above
- * copyright notice and this permission notice appear in all copies.
+ * Copyright (c) Robert Clausecker <f...@freebsd.org>
+ * Based on a publication by Daniel Lemire.
+ * Public domain where applicable.
*
- * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
- * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
- * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
- * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
- * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
- * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ * Daniel Lemire, "Fast Random Integer Generation in an Interval",
+ * Association for Computing Machinery, ACM Trans. Model. Comput. Simul.,
+ * no. 1, vol. 29, pp. 1--12, New York, NY, USA, January 2019.
*/
#include <stdint.h>
#include <stdlib.h>
-/*
- * Calculate a uniformly distributed random number less than upper_bound
- * avoiding "modulo bias".
- *
- * Uniformity is achieved by generating new random numbers until the one
- * returned is outside the range [0, 2**32 % upper_bound). This
- * guarantees the selected random number will be inside
- * [2**32 % upper_bound, 2**32) which maps back to [0, upper_bound)
- * after reduction modulo upper_bound.
- */
uint32_t
arc4random_uniform(uint32_t upper_bound)
{
- uint32_t r, min;
-
- if (upper_bound < 2)
- return 0;
-
- /* 2**32 % x == (2**32 - x) % x */
- min = -upper_bound % upper_bound;
+ uint64_t product;
/*
- * This could theoretically loop forever but each retry has
- * p > 0.5 (worst case, usually far better) of selecting a
- * number inside the range we need, so it should rarely need
- * to re-roll.
+ * The paper uses these variable names:
+ *
+ * L -- log2(UINT32_MAX+1)
+ * s -- upper_bound
+ * x -- arc4random() return value
+ * m -- product
+ * l -- (uint32_t)product
+ * t -- threshold
*/
- for (;;) {
- r = arc4random();
- if (r >= min)
- break;
+
+ if (upper_bound <= 1)
+ return (0);
+
+ product = upper_bound * (uint64_t)arc4random();
+
+ if ((uint32_t)product < upper_bound) {
+ uint32_t threshold;
+
+ /* threshold = (2**32 - upper_bound) % upper_bound */
+ threshold = -upper_bound % upper_bound;
+ while ((uint32_t)product < threshold)
+ product = upper_bound * (uint64_t)arc4random();
}
- return r % upper_bound;
+ return (product >> 32);
}
